Prompt-Based Exemplar Super-Compression and Regeneration for Class-Incremental Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Duan, Ruxiao, Chen, Jieneng, Kortylewski, Adam, Yuille, Alan, Liu, Yaoyao
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913966566932480
author Duan, Ruxiao
Chen, Jieneng
Kortylewski, Adam
Yuille, Alan
Liu, Yaoyao
author_facet Duan, Ruxiao
Chen, Jieneng
Kortylewski, Adam
Yuille, Alan
Liu, Yaoyao
contents Replay-based methods in class-incremental learning (CIL) have attained remarkable success. Despite their effectiveness, the inherent memory restriction results in saving a limited number of exemplars with poor diversity. In this paper, we introduce PESCR, a novel approach that substantially increases the quantity and enhances the diversity of exemplars based on a pre-trained general-purpose diffusion model, without fine-tuning it on target datasets or storing it in the memory buffer. Images are compressed into visual and textual prompts, which are saved instead of the original images, decreasing memory consumption by a factor of 24. In subsequent phases, diverse exemplars are regenerated by the diffusion model. We further propose partial compression and diffusion-based data augmentation to minimize the domain gap between generated exemplars and real images. PESCR significantly improves CIL performance across multiple benchmarks, e.g., 3.2% above the previous state-of-the-art on ImageNet-100.
format Preprint
id arxiv_https___arxiv_org_abs_2311_18266
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Prompt-Based Exemplar Super-Compression and Regeneration for Class-Incremental Learning
Duan, Ruxiao
Chen, Jieneng
Kortylewski, Adam
Yuille, Alan
Liu, Yaoyao
Computer Vision and Pattern Recognition
Replay-based methods in class-incremental learning (CIL) have attained remarkable success. Despite their effectiveness, the inherent memory restriction results in saving a limited number of exemplars with poor diversity. In this paper, we introduce PESCR, a novel approach that substantially increases the quantity and enhances the diversity of exemplars based on a pre-trained general-purpose diffusion model, without fine-tuning it on target datasets or storing it in the memory buffer. Images are compressed into visual and textual prompts, which are saved instead of the original images, decreasing memory consumption by a factor of 24. In subsequent phases, diverse exemplars are regenerated by the diffusion model. We further propose partial compression and diffusion-based data augmentation to minimize the domain gap between generated exemplars and real images. PESCR significantly improves CIL performance across multiple benchmarks, e.g., 3.2% above the previous state-of-the-art on ImageNet-100.
title Prompt-Based Exemplar Super-Compression and Regeneration for Class-Incremental Learning
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2311.18266